DeepFakesON-Phys: DeepFakes Detection based on Heart Rate Estimation

View on arXiv ← Back to list

Authors: Javier Hernandez-Ortega, Ruben Tolosana, Julian Fierrez, Aythami Morales

Published: 2020-10-01 13:37:58+00:00

AI Summary

This paper proposes DeepFakesON-Phys, a deepfake detection framework that leverages remote photoplethysmography (rPPG) to analyze heart rate information from videos. By using a Convolutional Attention Network (CAN), it achieves over 98% AUC on Celeb-DF and DFDC datasets, outperforming state-of-the-art methods.

Abstract

This work introduces a novel DeepFake detection framework based on physiological measurement. In particular, we consider information related to the heart rate using remote photoplethysmography (rPPG). rPPG methods analyze video sequences looking for subtle color changes in the human skin, revealing the presence of human blood under the tissues. In this work we investigate to what extent rPPG is useful for the detection of DeepFake videos. The proposed fake detector named DeepFakesON-Phys uses a Convolutional Attention Network (CAN), which extracts spatial and temporal information from video frames, analyzing and combining both sources to better detect fake videos. This detection approach has been experimentally evaluated using the latest public databases in the field: Celeb-DF and DFDC. The results achieved, above 98% AUC (Area Under the Curve) on both databases, outperform the state of the art and prove the success of fake detectors based on physiological measurement to detect the latest DeepFake videos.

Key findings

DeepFakesON-Phys achieves above 98% AUC on both Celeb-DF v2 and DFDC Preview datasets. The results demonstrate the effectiveness of using physiological measurements (rPPG) for deepfake detection and suggest that current deepfake methods do not adequately replicate physiological signals. The approach outperforms existing state-of-the-art methods.

Approach

DeepFakesON-Phys uses a Convolutional Attention Network (CAN) with two branches (Motion and Appearance Models) to extract spatio-temporal information from video frames. It leverages rPPG signals, analyzing subtle color changes in the skin to detect inconsistencies in heart rate indicative of deepfakes. The model is fine-tuned using pre-trained weights from a heart rate estimation model.

Datasets

Celeb-DF v2, DFDC Preview

Model(s)

Convolutional Attention Network (CAN)

Author countries

Spain

← Previous